1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * CUSE: Character device in Userspace
4 *
5 * Copyright (C) 2008-2009  SUSE Linux Products GmbH
6 * Copyright (C) 2008-2009  Tejun Heo <tj@kernel.org>
7 *
8 * CUSE enables character devices to be implemented from userland much
9 * like FUSE allows filesystems.  On initialization /dev/cuse is
10 * created.  By opening the file and replying to the CUSE_INIT request
11 * userland CUSE server can create a character device.  After that the
12 * operation is very similar to FUSE.
13 *
14 * A CUSE instance involves the following objects.
15 *
16 * cuse_conn	: contains fuse_conn and serves as bonding structure
17 * channel	: file handle connected to the userland CUSE server
18 * cdev		: the implemented character device
19 * dev		: generic device for cdev
20 *
21 * Note that 'channel' is what 'dev' is in FUSE.  As CUSE deals with
22 * devices, it's called 'channel' to reduce confusion.
23 *
24 * channel determines when the character device dies.  When channel is
25 * closed, everything begins to destruct.  The cuse_conn is taken off
26 * the lookup table preventing further access from cdev, cdev and
27 * generic device are removed and the base reference of cuse_conn is
28 * put.
29 *
30 * On each open, the matching cuse_conn is looked up and if found an
31 * additional reference is taken which is released when the file is
32 * closed.
33 */
34
35#define pr_fmt(fmt) "CUSE: " fmt
36
37#include <linux/fuse.h>
38#include <linux/cdev.h>
39#include <linux/device.h>
40#include <linux/file.h>
41#include <linux/fs.h>
42#include <linux/kdev_t.h>
43#include <linux/kthread.h>
44#include <linux/list.h>
45#include <linux/magic.h>
46#include <linux/miscdevice.h>
47#include <linux/mutex.h>
48#include <linux/slab.h>
49#include <linux/stat.h>
50#include <linux/module.h>
51#include <linux/uio.h>
52#include <linux/user_namespace.h>
53
54#include "fuse_i.h"
55
56#define CUSE_CONNTBL_LEN	64
57
58struct cuse_conn {
59	struct list_head	list;	/* linked on cuse_conntbl */
60	struct fuse_mount	fm;	/* Dummy mount referencing fc */
61	struct fuse_conn	fc;	/* fuse connection */
62	struct cdev		*cdev;	/* associated character device */
63	struct device		*dev;	/* device representing @cdev */
64
65	/* init parameters, set once during initialization */
66	bool			unrestricted_ioctl;
67};
68
69static DEFINE_MUTEX(cuse_lock);		/* protects registration */
70static struct list_head cuse_conntbl[CUSE_CONNTBL_LEN];
71static struct class *cuse_class;
72
73static struct cuse_conn *fc_to_cc(struct fuse_conn *fc)
74{
75	return container_of(fc, struct cuse_conn, fc);
76}
77
78static struct list_head *cuse_conntbl_head(dev_t devt)
79{
80	return &cuse_conntbl[(MAJOR(devt) + MINOR(devt)) % CUSE_CONNTBL_LEN];
81}
82
83
84/**************************************************************************
85 * CUSE frontend operations
86 *
87 * These are file operations for the character device.
88 *
89 * On open, CUSE opens a file from the FUSE mnt and stores it to
90 * private_data of the open file.  All other ops call FUSE ops on the
91 * FUSE file.
92 */
93
94static ssize_t cuse_read_iter(struct kiocb *kiocb, struct iov_iter *to)
95{
96	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(kiocb);
97	loff_t pos = 0;
98
99	return fuse_direct_io(&io, to, &pos, FUSE_DIO_CUSE);
100}
101
102static ssize_t cuse_write_iter(struct kiocb *kiocb, struct iov_iter *from)
103{
104	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(kiocb);
105	loff_t pos = 0;
106	/*
107	 * No locking or generic_write_checks(), the server is
108	 * responsible for locking and sanity checks.
109	 */
110	return fuse_direct_io(&io, from, &pos,
111			      FUSE_DIO_WRITE | FUSE_DIO_CUSE);
112}
113
114static int cuse_open(struct inode *inode, struct file *file)
115{
116	dev_t devt = inode->i_cdev->dev;
117	struct cuse_conn *cc = NULL, *pos;
118	int rc;
119
120	/* look up and get the connection */
121	mutex_lock(&cuse_lock);
122	list_for_each_entry(pos, cuse_conntbl_head(devt), list)
123		if (pos->dev->devt == devt) {
124			fuse_conn_get(&pos->fc);
125			cc = pos;
126			break;
127		}
128	mutex_unlock(&cuse_lock);
129
130	/* dead? */
131	if (!cc)
132		return -ENODEV;
133
134	/*
135	 * Generic permission check is already done against the chrdev
136	 * file, proceed to open.
137	 */
138	rc = fuse_do_open(&cc->fm, 0, file, 0);
139	if (rc)
140		fuse_conn_put(&cc->fc);
141	return rc;
142}
143
144static int cuse_release(struct inode *inode, struct file *file)
145{
146	struct fuse_file *ff = file->private_data;
147	struct fuse_mount *fm = ff->fm;
148
149	fuse_sync_release(NULL, ff, file->f_flags);
150	fuse_conn_put(fm->fc);
151
152	return 0;
153}
154
155static long cuse_file_ioctl(struct file *file, unsigned int cmd,
156			    unsigned long arg)
157{
158	struct fuse_file *ff = file->private_data;
159	struct cuse_conn *cc = fc_to_cc(ff->fm->fc);
160	unsigned int flags = 0;
161
162	if (cc->unrestricted_ioctl)
163		flags |= FUSE_IOCTL_UNRESTRICTED;
164
165	return fuse_do_ioctl(file, cmd, arg, flags);
166}
167
168static long cuse_file_compat_ioctl(struct file *file, unsigned int cmd,
169				   unsigned long arg)
170{
171	struct fuse_file *ff = file->private_data;
172	struct cuse_conn *cc = fc_to_cc(ff->fm->fc);
173	unsigned int flags = FUSE_IOCTL_COMPAT;
174
175	if (cc->unrestricted_ioctl)
176		flags |= FUSE_IOCTL_UNRESTRICTED;
177
178	return fuse_do_ioctl(file, cmd, arg, flags);
179}
180
181static const struct file_operations cuse_frontend_fops = {
182	.owner			= THIS_MODULE,
183	.read_iter		= cuse_read_iter,
184	.write_iter		= cuse_write_iter,
185	.open			= cuse_open,
186	.release		= cuse_release,
187	.unlocked_ioctl		= cuse_file_ioctl,
188	.compat_ioctl		= cuse_file_compat_ioctl,
189	.poll			= fuse_file_poll,
190	.llseek		= noop_llseek,
191};
192
193
194/**************************************************************************
195 * CUSE channel initialization and destruction
196 */
197
198struct cuse_devinfo {
199	const char		*name;
200};
201
202/**
203 * cuse_parse_one - parse one key=value pair
204 * @pp: i/o parameter for the current position
205 * @end: points to one past the end of the packed string
206 * @keyp: out parameter for key
207 * @valp: out parameter for value
208 *
209 * *@pp points to packed strings - "key0=val0\0key1=val1\0" which ends
210 * at @end - 1.  This function parses one pair and set *@keyp to the
211 * start of the key and *@valp to the start of the value.  Note that
212 * the original string is modified such that the key string is
213 * terminated with '\0'.  *@pp is updated to point to the next string.
214 *
215 * RETURNS:
216 * 1 on successful parse, 0 on EOF, -errno on failure.
217 */
218static int cuse_parse_one(char **pp, char *end, char **keyp, char **valp)
219{
220	char *p = *pp;
221	char *key, *val;
222
223	while (p < end && *p == '\0')
224		p++;
225	if (p == end)
226		return 0;
227
228	if (end[-1] != '\0') {
229		pr_err("info not properly terminated\n");
230		return -EINVAL;
231	}
232
233	key = val = p;
234	p += strlen(p);
235
236	if (valp) {
237		strsep(&val, "=");
238		if (!val)
239			val = key + strlen(key);
240		key = strstrip(key);
241		val = strstrip(val);
242	} else
243		key = strstrip(key);
244
245	if (!strlen(key)) {
246		pr_err("zero length info key specified\n");
247		return -EINVAL;
248	}
249
250	*pp = p;
251	*keyp = key;
252	if (valp)
253		*valp = val;
254
255	return 1;
256}
257
258/**
259 * cuse_parse_devinfo - parse device info
260 * @p: device info string
261 * @len: length of device info string
262 * @devinfo: out parameter for parsed device info
263 *
264 * Parse @p to extract device info and store it into @devinfo.  String
265 * pointed to by @p is modified by parsing and @devinfo points into
266 * them, so @p shouldn't be freed while @devinfo is in use.
267 *
268 * RETURNS:
269 * 0 on success, -errno on failure.
270 */
271static int cuse_parse_devinfo(char *p, size_t len, struct cuse_devinfo *devinfo)
272{
273	char *end = p + len;
274	char *key, *val;
275	int rc;
276
277	while (true) {
278		rc = cuse_parse_one(&p, end, &key, &val);
279		if (rc < 0)
280			return rc;
281		if (!rc)
282			break;
283		if (strcmp(key, "DEVNAME") == 0)
284			devinfo->name = val;
285		else
286			pr_warn("unknown device info \"%s\"\n", key);
287	}
288
289	if (!devinfo->name || !strlen(devinfo->name)) {
290		pr_err("DEVNAME unspecified\n");
291		return -EINVAL;
292	}
293
294	return 0;
295}
296
297static void cuse_gendev_release(struct device *dev)
298{
299	kfree(dev);
300}
301
302struct cuse_init_args {
303	struct fuse_args_pages ap;
304	struct cuse_init_in in;
305	struct cuse_init_out out;
306	struct page *page;
307	struct fuse_page_desc desc;
308};
309
310/**
311 * cuse_process_init_reply - finish initializing CUSE channel
312 *
313 * This function creates the character device and sets up all the
314 * required data structures for it.  Please read the comment at the
315 * top of this file for high level overview.
316 */
317static void cuse_process_init_reply(struct fuse_mount *fm,
318				    struct fuse_args *args, int error)
319{
320	struct fuse_conn *fc = fm->fc;
321	struct cuse_init_args *ia = container_of(args, typeof(*ia), ap.args);
322	struct fuse_args_pages *ap = &ia->ap;
323	struct cuse_conn *cc = fc_to_cc(fc), *pos;
324	struct cuse_init_out *arg = &ia->out;
325	struct page *page = ap->pages[0];
326	struct cuse_devinfo devinfo = { };
327	struct device *dev;
328	struct cdev *cdev;
329	dev_t devt;
330	int rc, i;
331
332	if (error || arg->major != FUSE_KERNEL_VERSION || arg->minor < 11)
333		goto err;
334
335	fc->minor = arg->minor;
336	fc->max_read = max_t(unsigned, arg->max_read, 4096);
337	fc->max_write = max_t(unsigned, arg->max_write, 4096);
338
339	/* parse init reply */
340	cc->unrestricted_ioctl = arg->flags & CUSE_UNRESTRICTED_IOCTL;
341
342	rc = cuse_parse_devinfo(page_address(page), ap->args.out_args[1].size,
343				&devinfo);
344	if (rc)
345		goto err;
346
347	/* determine and reserve devt */
348	devt = MKDEV(arg->dev_major, arg->dev_minor);
349	if (!MAJOR(devt))
350		rc = alloc_chrdev_region(&devt, MINOR(devt), 1, devinfo.name);
351	else
352		rc = register_chrdev_region(devt, 1, devinfo.name);
353	if (rc) {
354		pr_err("failed to register chrdev region\n");
355		goto err;
356	}
357
358	/* devt determined, create device */
359	rc = -ENOMEM;
360	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
361	if (!dev)
362		goto err_region;
363
364	device_initialize(dev);
365	dev_set_uevent_suppress(dev, 1);
366	dev->class = cuse_class;
367	dev->devt = devt;
368	dev->release = cuse_gendev_release;
369	dev_set_drvdata(dev, cc);
370	dev_set_name(dev, "%s", devinfo.name);
371
372	mutex_lock(&cuse_lock);
373
374	/* make sure the device-name is unique */
375	for (i = 0; i < CUSE_CONNTBL_LEN; ++i) {
376		list_for_each_entry(pos, &cuse_conntbl[i], list)
377			if (!strcmp(dev_name(pos->dev), dev_name(dev)))
378				goto err_unlock;
379	}
380
381	rc = device_add(dev);
382	if (rc)
383		goto err_unlock;
384
385	/* register cdev */
386	rc = -ENOMEM;
387	cdev = cdev_alloc();
388	if (!cdev)
389		goto err_unlock;
390
391	cdev->owner = THIS_MODULE;
392	cdev->ops = &cuse_frontend_fops;
393
394	rc = cdev_add(cdev, devt, 1);
395	if (rc)
396		goto err_cdev;
397
398	cc->dev = dev;
399	cc->cdev = cdev;
400
401	/* make the device available */
402	list_add(&cc->list, cuse_conntbl_head(devt));
403	mutex_unlock(&cuse_lock);
404
405	/* announce device availability */
406	dev_set_uevent_suppress(dev, 0);
407	kobject_uevent(&dev->kobj, KOBJ_ADD);
408out:
409	kfree(ia);
410	__free_page(page);
411	return;
412
413err_cdev:
414	cdev_del(cdev);
415err_unlock:
416	mutex_unlock(&cuse_lock);
417	put_device(dev);
418err_region:
419	unregister_chrdev_region(devt, 1);
420err:
421	fuse_abort_conn(fc);
422	goto out;
423}
424
425static int cuse_send_init(struct cuse_conn *cc)
426{
427	int rc;
428	struct page *page;
429	struct fuse_mount *fm = &cc->fm;
430	struct cuse_init_args *ia;
431	struct fuse_args_pages *ap;
432
433	BUILD_BUG_ON(CUSE_INIT_INFO_MAX > PAGE_SIZE);
434
435	rc = -ENOMEM;
436	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
437	if (!page)
438		goto err;
439
440	ia = kzalloc(sizeof(*ia), GFP_KERNEL);
441	if (!ia)
442		goto err_free_page;
443
444	ap = &ia->ap;
445	ia->in.major = FUSE_KERNEL_VERSION;
446	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
447	ia->in.flags |= CUSE_UNRESTRICTED_IOCTL;
448	ap->args.opcode = CUSE_INIT;
449	ap->args.in_numargs = 1;
450	ap->args.in_args[0].size = sizeof(ia->in);
451	ap->args.in_args[0].value = &ia->in;
452	ap->args.out_numargs = 2;
453	ap->args.out_args[0].size = sizeof(ia->out);
454	ap->args.out_args[0].value = &ia->out;
455	ap->args.out_args[1].size = CUSE_INIT_INFO_MAX;
456	ap->args.out_argvar = true;
457	ap->args.out_pages = true;
458	ap->num_pages = 1;
459	ap->pages = &ia->page;
460	ap->descs = &ia->desc;
461	ia->page = page;
462	ia->desc.length = ap->args.out_args[1].size;
463	ap->args.end = cuse_process_init_reply;
464
465	rc = fuse_simple_background(fm, &ap->args, GFP_KERNEL);
466	if (rc) {
467		kfree(ia);
468err_free_page:
469		__free_page(page);
470	}
471err:
472	return rc;
473}
474
475static void cuse_fc_release(struct fuse_conn *fc)
476{
477	kfree(fc_to_cc(fc));
478}
479
480/**
481 * cuse_channel_open - open method for /dev/cuse
482 * @inode: inode for /dev/cuse
483 * @file: file struct being opened
484 *
485 * Userland CUSE server can create a CUSE device by opening /dev/cuse
486 * and replying to the initialization request kernel sends.  This
487 * function is responsible for handling CUSE device initialization.
488 * Because the fd opened by this function is used during
489 * initialization, this function only creates cuse_conn and sends
490 * init.  The rest is delegated to a kthread.
491 *
492 * RETURNS:
493 * 0 on success, -errno on failure.
494 */
495static int cuse_channel_open(struct inode *inode, struct file *file)
496{
497	struct fuse_dev *fud;
498	struct cuse_conn *cc;
499	int rc;
500
501	/* set up cuse_conn */
502	cc = kzalloc(sizeof(*cc), GFP_KERNEL);
503	if (!cc)
504		return -ENOMEM;
505
506	/*
507	 * Limit the cuse channel to requests that can
508	 * be represented in file->f_cred->user_ns.
509	 */
510	fuse_conn_init(&cc->fc, &cc->fm, file->f_cred->user_ns,
511		       &fuse_dev_fiq_ops, NULL);
512
513	cc->fc.release = cuse_fc_release;
514	fud = fuse_dev_alloc_install(&cc->fc);
515	fuse_conn_put(&cc->fc);
516	if (!fud)
517		return -ENOMEM;
518
519	INIT_LIST_HEAD(&cc->list);
520
521	cc->fc.initialized = 1;
522	rc = cuse_send_init(cc);
523	if (rc) {
524		fuse_dev_free(fud);
525		return rc;
526	}
527	file->private_data = fud;
528
529	return 0;
530}
531
532/**
533 * cuse_channel_release - release method for /dev/cuse
534 * @inode: inode for /dev/cuse
535 * @file: file struct being closed
536 *
537 * Disconnect the channel, deregister CUSE device and initiate
538 * destruction by putting the default reference.
539 *
540 * RETURNS:
541 * 0 on success, -errno on failure.
542 */
543static int cuse_channel_release(struct inode *inode, struct file *file)
544{
545	struct fuse_dev *fud = file->private_data;
546	struct cuse_conn *cc = fc_to_cc(fud->fc);
547
548	/* remove from the conntbl, no more access from this point on */
549	mutex_lock(&cuse_lock);
550	list_del_init(&cc->list);
551	mutex_unlock(&cuse_lock);
552
553	/* remove device */
554	if (cc->dev)
555		device_unregister(cc->dev);
556	if (cc->cdev) {
557		unregister_chrdev_region(cc->cdev->dev, 1);
558		cdev_del(cc->cdev);
559	}
560
561	return fuse_dev_release(inode, file);
562}
563
564static struct file_operations cuse_channel_fops; /* initialized during init */
565
566
567/**************************************************************************
568 * Misc stuff and module initializatiion
569 *
570 * CUSE exports the same set of attributes to sysfs as fusectl.
571 */
572
573static ssize_t cuse_class_waiting_show(struct device *dev,
574				       struct device_attribute *attr, char *buf)
575{
576	struct cuse_conn *cc = dev_get_drvdata(dev);
577
578	return sprintf(buf, "%d\n", atomic_read(&cc->fc.num_waiting));
579}
580static DEVICE_ATTR(waiting, 0400, cuse_class_waiting_show, NULL);
581
582static ssize_t cuse_class_abort_store(struct device *dev,
583				      struct device_attribute *attr,
584				      const char *buf, size_t count)
585{
586	struct cuse_conn *cc = dev_get_drvdata(dev);
587
588	fuse_abort_conn(&cc->fc);
589	return count;
590}
591static DEVICE_ATTR(abort, 0200, NULL, cuse_class_abort_store);
592
593static struct attribute *cuse_class_dev_attrs[] = {
594	&dev_attr_waiting.attr,
595	&dev_attr_abort.attr,
596	NULL,
597};
598ATTRIBUTE_GROUPS(cuse_class_dev);
599
600static struct miscdevice cuse_miscdev = {
601	.minor		= CUSE_MINOR,
602	.name		= "cuse",
603	.fops		= &cuse_channel_fops,
604};
605
606MODULE_ALIAS_MISCDEV(CUSE_MINOR);
607MODULE_ALIAS("devname:cuse");
608
609static int __init cuse_init(void)
610{
611	int i, rc;
612
613	/* init conntbl */
614	for (i = 0; i < CUSE_CONNTBL_LEN; i++)
615		INIT_LIST_HEAD(&cuse_conntbl[i]);
616
617	/* inherit and extend fuse_dev_operations */
618	cuse_channel_fops		= fuse_dev_operations;
619	cuse_channel_fops.owner		= THIS_MODULE;
620	cuse_channel_fops.open		= cuse_channel_open;
621	cuse_channel_fops.release	= cuse_channel_release;
622	/* CUSE is not prepared for FUSE_DEV_IOC_CLONE */
623	cuse_channel_fops.unlocked_ioctl	= NULL;
624
625	cuse_class = class_create("cuse");
626	if (IS_ERR(cuse_class))
627		return PTR_ERR(cuse_class);
628
629	cuse_class->dev_groups = cuse_class_dev_groups;
630
631	rc = misc_register(&cuse_miscdev);
632	if (rc) {
633		class_destroy(cuse_class);
634		return rc;
635	}
636
637	return 0;
638}
639
640static void __exit cuse_exit(void)
641{
642	misc_deregister(&cuse_miscdev);
643	class_destroy(cuse_class);
644}
645
646module_init(cuse_init);
647module_exit(cuse_exit);
648
649MODULE_AUTHOR("Tejun Heo <tj@kernel.org>");
650MODULE_DESCRIPTION("Character device in Userspace");
651MODULE_LICENSE("GPL");
652